Sending OpenAI Chat Completion Traces
This guide explains how to send OpenTelemetry-compatible traces for OpenAI chat completions to Langtrace using cURL.
Here’s an example of how your traces will appear in the Langtrace UI after following this guide:
Prerequisites
- Langtrace API Key
- OpenAI API Key (for comparison with SDK usage)
Endpoint
POST http://localhost:3000/api/trace # For local development
# or
POST https://api.langtrace.ai/api/trace # For production
Content-Type: application/json
x-api-key: YOUR_LANGTRACE_API_KEY
User-Agent: opentelemetry-python
The trace must follow the OpenTelemetry format and include specific attributes for OpenAI chat completions. Note that attributes must use the OpenTelemetry value format with stringValue or intValue:
{
"resourceSpans": [{
"resource": {
"attributes": [{
"key": "service.name",
"value": { "stringValue": "your-service-name" }
}]
},
"scopeSpans": [{
"scope": {
"name": "openai",
"version": "1.0.0"
},
"spans": [{
"name": "OpenAI ChatCompletion",
"kind": 2, // SpanKind.CLIENT
"startTimeUnixNano": "1234567890000000000",
"endTimeUnixNano": "1234567891000000000",
"traceId": "a1b2c3d4e5f6g7h8i9j0k1l2m3n4o5p6",
"spanId": "a1b2c3d4e5f6g7h8",
"attributes": [{
"key": "llm.vendor",
"value": { "stringValue": "openai" }
}, {
"key": "llm.request.model",
"value": { "stringValue": "gpt-3.5-turbo" }
}, {
"key": "llm.request.messages",
"value": { "stringValue": "[{\"role\": \"user\", \"content\": \"Hello\"}]" }
}, {
"key": "llm.path",
"value": { "stringValue": "/v1/chat/completions" }
}, {
"key": "llm.request.temperature",
"value": { "intValue": 0.7 }
}, {
"key": "llm.request.max_tokens",
"value": { "intValue": 150 }
}, {
"key": "llm.usage.prompt_tokens",
"value": { "intValue": 10 }
}, {
"key": "llm.usage.completion_tokens",
"value": { "intValue": 15 }
}, {
"key": "llm.usage.total_tokens",
"value": { "intValue": 25 }
}]
}]
}]
}]
}
Example cURL Command
curl -X POST "http://localhost:3000/api/trace" \
-H "Content-Type: application/json" \
-H "x-api-key: ${LANGTRACE_API_KEY}" \
-H "User-Agent: opentelemetry-python" \
-d '{
"resourceSpans": [{
"resource": {
"attributes": [{
"key": "service.name",
"value": { "stringValue": "openai-test" }
}]
},
"scopeSpans": [{
"scope": {
"name": "openai",
"version": "1.0.0"
},
"spans": [{
"name": "OpenAI ChatCompletion",
"kind": 2,
"startTimeUnixNano": "'$(date +%s%N)'",
"endTimeUnixNano": "'$(date +%s%N)'",
"traceId": "'$(openssl rand -hex 16)'",
"spanId": "'$(openssl rand -hex 8)'",
"attributes": [{
"key": "llm.vendor",
"value": { "stringValue": "openai" }
}, {
"key": "llm.request.model",
"value": { "stringValue": "gpt-3.5-turbo" }
}, {
"key": "llm.request.messages",
"value": { "stringValue": "[{\"role\": \"user\", \"content\": \"Hello\"}]" }
}, {
"key": "llm.path",
"value": { "stringValue": "/v1/chat/completions" }
}, {
"key": "llm.request.temperature",
"value": { "intValue": 0.7 }
}, {
"key": "llm.request.max_tokens",
"value": { "intValue": 150 }
}, {
"key": "llm.usage.prompt_tokens",
"value": { "intValue": 10 }
}, {
"key": "llm.usage.completion_tokens",
"value": { "intValue": 15 }
}, {
"key": "llm.usage.total_tokens",
"value": { "intValue": 25 }
}]
}]
}]
}]
}'
Required Attributes
| Attribute | Description | Required |
|---|
| llm.vendor | Service provider (“openai”) | Yes |
| llm.request.model | Model name (e.g., “gpt-3.5-turbo”) | Yes |
| llm.request.messages | JSON string of messages array | Yes |
| llm.path | API endpoint path | Yes |
| llm.request.temperature | Temperature setting | No |
| llm.request.max_tokens | Maximum tokens | No |
| llm.usage.* | Token usage statistics | No |
Function Calling Example
When using OpenAI function calling, include the tools in the attributes with proper value formatting:
curl -X POST "http://localhost:3000/api/trace" \
-H "Content-Type: application/json" \
-H "x-api-key: ${LANGTRACE_API_KEY}" \
-H "User-Agent: opentelemetry-python" \
-d '{
"resourceSpans": [{
"resource": {
"attributes": [{
"key": "service.name",
"value": { "stringValue": "openai-test" }
}]
},
"scopeSpans": [{
"scope": {
"name": "openai",
"version": "1.0.0"
},
"spans": [{
"name": "OpenAI ChatCompletion",
"kind": 2,
"startTimeUnixNano": "'$(date +%s%N)'",
"endTimeUnixNano": "'$(date +%s%N)'",
"traceId": "'$(openssl rand -hex 16)'",
"spanId": "'$(openssl rand -hex 8)'",
"attributes": [{
"key": "llm.vendor",
"value": { "stringValue": "openai" }
}, {
"key": "llm.request.model",
"value": { "stringValue": "gpt-3.5-turbo" }
}, {
"key": "llm.request.messages",
"value": { "stringValue": "[{\"role\": \"user\", \"content\": \"What's the weather?\"}]" }
}, {
"key": "llm.path",
"value": { "stringValue": "/v1/chat/completions" }
}, {
"key": "llm.tools",
"value": { "stringValue": "[{\"type\": \"function\", \"function\": {\"name\": \"get_weather\", \"arguments\": \"{\\\"location\\\": \\\"London\\\"}\"}}]" }
}]
}]
}]
}]
}'
Success Response
{
"message": "Traces added successfully"
}
Error Response
{
"name": "Error name",
"message": "Error message",
"stack": "Error stack trace",
"fullError": "Detailed error information"
}
Comparison with SDK Usage
For comparison, here’s how the same trace would be generated using the Python SDK:
from openai import OpenAI
from langtrace_python_sdk import langtrace
# Initialize clients
client = OpenAI()
langtrace.init(api_key="YOUR_LANGTRACE_API_KEY")
# Make API call - traces are automatically sent
response = client.chat.completions.create(
model="gpt-3.5-turbo",
messages=[{"role": "user", "content": "Hello"}],
temperature=0.7,
max_tokens=150
)
Best Practices
- Always include required attributes (vendor, model, messages, path)
- Generate unique trace and span IDs for each request
- Use accurate timestamps for start and end times
- Include usage statistics when available
- Format message arrays and tool calls as proper JSON strings
- Use the OpenTelemetry user agent to ensure proper trace processing
- Include resource attributes with service information
- Use proper OpenTelemetry value format (stringValue/intValue) for attributes
